Professional Documents
Culture Documents
Reading Materials
Reading Materials
Gary Richardson
University of California, Irvine and National Bureau of Economic Research
William Troost
University of Southern California
The Federal Reserve Act divided Mississippi between the 6th (Atlanta)
and 8th (St. Louis) Districts. During the Great Depression, these dis-
tricts’ policies differed. Atlanta championed monetary activism and
the extension of aid to ailing banks. St. Louis eschewed expansionary
initiatives. During a banking crisis in 1930, Atlanta expedited lending
to banks in need. St. Louis did not. Outcomes differed across districts.
In Atlanta, banks survived at higher rates, lending continued at higher
levels, commerce contracted less, and recovery began earlier. These
patterns indicate that central bank intervention influenced bank
health, credit availability, and business activity.
We thank friends and colleagues for advice and encouragement. Dan Bogart, Michael
Bordo, Jan Brueckner, Charles Calomiris, Michelle Garfinkel, Joseph Mason, Christina
Romer, David Romer, Jean-Laurent Rosenthal, Eugene White, an anonymous referee, and
participants in seminars at the Federal Deposit Insurance Corporation, Federal Reserve
Bank of Atlanta, Federal Reserve Bank of Chicago, Federal Reserve Bank of St. Louis,
Federal Reserve Board of Governors, NBER Summer Institute, Rutgers University, George
Mason University, University of California, Berkeley, University of California, Irvine, and
Western Economics Association provided comments on earlier drafts. This material is based
on work supported by the National Science Foundation under grant 0551232. Shagufta
Ahmed, Shaista Ahmed, Ching-Yi Chung, Joanna Do, and William Troost provided ex-
ceptional research assistance.
1031
TABLE 1
Number of Banks in Mississippi on July 1 of Each Year
(e.g., during the fall of 1931) and for other reasons (e.g., loan defaults)
shows little (or no) correlation with subsequent commercial contraction.
Our methods directly address key questions concerning the collapse
of the banking system during the early 1930s. Did Federal Reserve pol-
icies influence bank failure rates? Did providing liquidity (or credibly
committing to do so) reduce rates of bank suspension and liquidation?
Did the demise of banks reduce the supply of commercial credit? Did
the contraction of credit reduce commercial activity? To each of these
questions, the answer is yes.
Section VII discusses the implications of our analysis. By injecting
liquidity into the banking system, particularly during the banking panic
in the fall of 1930, the Federal Reserve Bank of Atlanta prevented bank
failures rates and facilitated business activity. If other Federal Reserve
banks had pursued similar strategies, fewer banks would have failed,
and the depression may have followed a different course.
6th Federal Reserve District (Atlanta) 8th Federal Reserve District (St. Louis)
All 6th (N p 141) Near Border (N p 76) Near Border (N p 169) All 8th N p 112)
Standard Standard Standard Standard
Median Mean Deviation Median Mean Deviation Median Mean Deviation Median Mean Deviation
Financial ratios:
Net worth/total assets .10 .11 .04 .10 .11 .04 .13 .14 .06 .11 .13 .05
Cash/total assets .37 .38 .14 .36 .39 .14 .38 .37 .15 .38 .38 .15
Deposits/total liabilities .87 .85 .07 .88 .85 .08 .85 .82 .11 .86 .83 .10
Financial characteristics:
Total assets ($1,000) 559 1,166 141 514 1,211 225 451 790 106 448 748 76
Loans and discounts ($1,000) 334 676 1,070 278 713 1,288 270 464 755 256 437 668
Cash and exchanges ($1,000) 92 204 310 84 228 373 92 174 276 91 157 237
Deposits ($1,000) 506 1,003 1,445 465 1,040 1,699 379 662 993 369 629 869
Paid-up capital ($1,000) 30 59 75 30 63 86 30 52 65 30 49 57
State-chartered banks (%) .85 .36 .88 .33 .90 .30 .92 .28
Federal Reserve member (%) .15 .36 .12 .33 .12 .32 .10 .30
Years in operation 24 23.2 12.3 24.5 24.0 12.7 21 21.9 14.9 20.5 21.8 14.2
Correspondents (N) 3 3.10 .90 3 3.08 .95 3 3.04 .89 3 2.96 .96
Source.—Rand McNally Bankers’ Directory, various July issues, 1929–35.
Note.—Near border sample consists of banks in counties for which at least 50 percent of the area lies within 1 degree latitude of the Federal Reserve district border.
A. Policy Regimes
Friedman and Schwartz (1963) pioneered efforts to identify Federal
Reserve policy regimes using the narrative historical approach. Romer
and Romer (1989) emphasize the importance of establishing clear cri-
teria for identifying policy regimes. Since our essay focuses on banking
crises, we define regimes in terms of policies regarding whether and
how to intervene during widespread and rapid failures of financial in-
stitutions that contemporaries classified as panics.
In the spring of 1913, the organizing committee of the Federal Re-
serve System split the state of Mississippi evenly between the 6th and
8th Districts. Figure 1 depicts the division (as well as information dis-
cussed later). In the 6th District from 1913 until the mid-1930s, the
Atlanta Fed followed Bagehot’s rule, a doctrine that during financial
panics, central banks should act as lenders of last resort and extend
credit to institutions afflicted by illiquidity. Such lending should be suf-
ficient to enable solvent but illiquid institutions to survive deposit losses
Fig. 1.—Mississippi’s division into Federal Reserve districts and bank suspensions be-
tween October 1930 and March 1931. Source: See Section II. The solid line represents
the Federal Reserve district border. The dotted lines enclose the counties for which at
least half the area lies within 1 degree latitude of the district border.
and, thus, to prevent runs from driving healthy banks into insolvency.
From 1913 to 1929, the Atlanta Fed faced four panics in which it could
employ such policies. In each instance, the Atlanta Fed rushed large
quantities of cash to the afflicted region, extended emergency loans to
member banks, helped member banks extend credit to their country
clients, and returned the situation to status quo ex ante (for details, see
Richardson and Troost [2006]). During the depression of the 1930s,
the Atlanta Fed consistently advocated monetary expansion. Atlanta’s
advocacy caught President Franklin Roosevelt’s attention, and he ap-
pointed Atlanta’s governor, Eugene Black, to be the chairman of the
Federal Reserve Board.
The policies of the Federal Reserve Bank of St. Louis were far dif-
B. Homogeneous Conditions
Mississippi was homogeneous economically and demographically. Table
3 demonstrates this by displaying county-level data drawn from the cen-
suses of population, manufacturing, and agriculture for 1930. In both
districts, the fraction of the population in the labor force was substantial.
Unemployment rates were low. Farm debt hovered around one-third to
one-fifth of farm value. Rural counties concentrated on cultivating cot-
ton. Prevailing prices for labor (average annual manufacturing wage)
and capital (ratio of interest charges to mortgage debt) differed little
across counties. The largest differences arose in the extremities of the
state. The counties adjoining the Federal Reserve district border had
few discernible differences.
Mississippi’s banking system was also homogeneous. The Banking De-
partment applied standard procedures throughout the state. So did the
Office of the Comptroller of Currency, Reconstruction Finance Cor-
poration, Department of Agriculture, Works Progress Administration,
and the intermediate credit banks, since Mississippi lay within a single
district for all these federal bureaus.
Fig. 2.—Discount response after the collapse of Caldwell, aggregate discounts each week
as a percentage of initial level. Source: See Section II.
total Federal Reserve credit to member banks in the 8th District declined
by more than $11,800,000 (Wicker 1996, 54).
Figure 2 illuminates changes in discount lending following Caldwell’s
collapse. Discounts of the 6th District rose rapidly to a peak 40 percent
higher than before the crisis. Discounts of the 8th District fell gradually,
as the extension of new loans slowed and existing loans expired.
In this situation, models of the Diamond-Dybvig type provide clear
predictions (Diamond and Dybvig 1983). Bank failure rates in the 6th
District should have been lower than bank failure rates in the 8th Dis-
trict, since a lender of last resort can mitigate financial panics by ex-
tending credit to illiquid institutions (and perhaps forestall a panic by
credibly committing to do so).3 Since the Atlanta Fed implemented such
a policy in a prompt, ample, and public manner, difference in outcomes
between the 6th and 8th Districts (if any) should reveal the effectiveness
(or ineffectiveness) of Atlanta’s policies.
Nonpanic periods serve as a control case that helps to test the ho-
mogeneity assumption underlying our analysis. Bagehot’s rule is a policy
implemented during panics, when withdrawals, contagion, and illiquid-
ity bedevil banks. The policy does not operate and therefore should
have no direct effect on bank failure rates outside of panic periods. The
period following the summer of 1931, when the St. Louis Fed adopted
3
Recent theoretical work indicates that monetary expansion can prevent the failure of
depository institutions, curtail contagion among banks, and alleviate the contraction of
liquidity even when real shocks (such as loan defaults or delays in bond repayment) are
the root of the problem and trigger the illiquidity crisis. See Diamond and Rajan (2005,
2006).
TABLE 4
Bank Suspensions and Liquidations
Fig. 3.—Percentage of banks in business and in operations in the 6th and 8th Federal
Reserve Districts in Mississippi, July 1929 to June 1933. Source: See Section II.
are banks whose doors are open to the public. Banks in business are
banks that are not bankrupt. The difference is the number of tempo-
rarily suspended banks. Figure 3 also indicates the date when the St.
Louis Fed’s policies began to converge toward those of the Atlanta Fed
and the dates of the events that the historical literature identifies as
triggers of the surges in suspensions apparent in the evidence. Figure
3 shows that during the post-Caldwell panic, when policy regimes dif-
fered across districts, banks suspended operations (temporarily and per-
manently) at much higher rates in the 8th District. During later surges
in bank suspensions, when policies differed little and the rise in failures
stemmed largely from fundamental factors, banks in the 6th and 8th
Districts failed at similar rates.
Figure 1 plots on a map of Mississippi each bank that suspends from
October 1930 through March 1931. The map indicates the division of
the state into the 6th (below solid line) and 8th (above solid line)
Districts. The area between the dotted lines indicates counties for which
at least half of the area lies within 1 degree latitude of the Federal
Reserve district border. The events precipitating the panic began in the
town that had five bank suspensions. The suspensions appear to be
distributed widely throughout both districts without prominent geo-
graphic patterns.
A. Nonparametric Estimates
The analysis of time to failure rests on survivor and hazard functions.
This subsection presents nonparametric estimates of survivor functions
constructed via the Kaplan-Meier method and of hazard functions con-
structed by smoothing raw hazard rates (i.e., the number of bank failures
divided by the number of banks at risk on each date). Kernels are
discrete Epanechnikov. Bandwidths of 28 days on graphs spanning 4
years and 7 days on graphs spanning 4 months are wide enough to
smooth daily volatility without obscuring weekly shifts in the probability
of failure.5
Figure 4 presents survival and hazard functions for all banks in Mis-
sissippi during the banking crisis in the fall of 1930. The time under
analysis is restricted to the four months following the collapse of Cald-
well and Company. The population at risk is all banks in operation. A
bank that surrendered its charter voluntarily or merged with another
institution departs from the population at risk (but is not counted as a
failure) on the date that it ceased operations. A bank that suspended
operations is counted as a failure on the date that it closed its doors to
the public.
5
Our estimates of the survival function, S(t) , the raw hazard function, h(t), and the
smoothed hazard function, g(t), are
Ŝ(t) p
ti!t
写
ni ⫺ di
ni
,
where ni is the number of banks in business at the beginning of time period ti , di is the
number of banks experiencing an event (such as entering receivership) at time ti , and
ti indicates the ith time period. The raw hazard for period ti is
di
ĥ(ti) p .
ni
The hazard function is estimated by smoothing raw hazards, so that the hazard in the ith
time period is
冘
u
ĝ(ti) p ˆ
Kzh(ti⫹z),
zp⫺u
6
Throughout this essay, whenever we state “within 1 degree latitude of the border,” we
are referring to this county-based distance definition. The set includes all banks operating
in a county for which at least 50 percent of the surface area lay within 1 degree latitude
of the border. This definition generates a band running through the center of the state
straddling the Federal Reserve district border. The outer edges of the band vary from 70
to 95 miles distance from the boundary.
1048
1049
1050
1051
B. Parametric Estimates
A plethora of potential parameterizations exist for our analysis. We pre-
sent results for the current gold standard in this literature, the log-
logistic survival model of Calomiris and Mason (2003). In this model,
the unit of observation is the individual bank. The dependent variable
is log days until distress. Time under observation begins on July 1, 1929,
and ends at the national banking holiday in March 1933. The explan-
atory variables include the characteristics of banks, the characteristics
of counties in which banks operate, a measure of business conditions
at the national level, indicators of periods of panic, and in our version
of this model, indicators of Federal Reserve policy regimes. Bank char-
acteristics update annually each July 1. County characteristics (from the
Census of 1930) remain constant over time. Economic conditions up-
date monthly. This framework allows us to determine the relative im-
portance of fundamentals and contagion as sources of bank distress and
to test whether Federal Reserve intervention mitigated (or accentuated)
banking panics.
Table 5 presents the results of this exercise. Column 1 reports the
basic model. It contains indicator variables for the three surges in bank
suspension in the fall of 1930, fall of 1931, and winter of 1933; for
whether a bank operated within the 6th District; and for whether during
each of three surges in bank suspensions a bank operated within the
6th District. The crisis indicators reveal to what extent failure rates rose
above the baseline during each surge. The crisis/district interaction
terms reveal for each crisis whether failure rates differed between the
6th and 8th Districts. The coefficients for the banking crises in 1930
and 1931 are statistically significant, indicating that during the crises,
the rate of bank distress rose above the baseline. The coefficient for
the fall 1930 crisis/Atlanta Fed interaction term is also statistically sig-
1054
(1.18) (.78) (.67) (.76) (1.35) (.53)
Fed Atlanta during crisis 1931 1.25 .42 .47 .31 .13 .28
(.84) (.65) (.59) (.64) (1.23) (.50)
Fed Atlanta during crisis 1933 .61 1.07 .97 .86 .73 .77
(.92) (.94) (.90) (.98) (1.35) (.72)
Federal Reserve Atlanta ⫺1.01* ⫺.90* ⫺.98* ⫺1.05* ⫺.93** ⫺.77*
(.38) (.36) (.47) (.43) (.52) (.33)
Banking crisis—fall 1930 ⫺12.38* ⫺3.76* ⫺3.50* ⫺3.70* ⫺4.19* ⫺2.67*
(1.19) (.92) (.77) (.92) (1.60) (.58)
Banking crisis—fall 1931 ⫺2.85* ⫺2.41* ⫺2.23* ⫺2.34* ⫺2.66* ⫺1.88*
(.74) (.59) (.54) (.56) (.78) (.45)
Banking crisis—winter 1933 ⫺1.00 ⫺1.51** ⫺1.40** ⫺1.50** ⫺1.46 ⫺.47
(.70) (.87) (.81) (.87) (1.33) (.56)
Assets % cash 6.37* 5.66* 6.31* 7.06* 4.55*
(1.51) (1.41) (1.53) (2.35) (1.142
1055
White sandwich method clustered on individual banks. CM indicates that the vector of control variables conforms to the specifications of Calomiris and Mason (2003). MS indicates that the vector
is fitted to Mississippi fundamentals. PC indicates that the vector consists of principal components of county variables. The border county sample consists of all banks located in counties for which
at least 50 percent of the area lay within 1 degree latitude of the Federal Reserve district border. The number of explanatory variables includes the constant and the curvature parameter.
* Significant at the 5 percent level.
** Significant at the 10 percent level.
TABLE 6
Magnitudes of Effects of Policy Regimes and Panics: Change in Cumulative
Hazard Rates in Log-Logistic Regressions
nificant, indicating that during the crisis, banks in the 6th District failed
at lower rates than banks in the 8th District. We cannot reject the null
hypothesis that the other coefficients equal zero.
Table 6 reveals the magnitudes of the coefficients. Column 1 indicates
that the crisis in the fall of 1930 raised bank failure rates substantially.
The marginal effects can be stated as changes in cumulative hazard rates
(a metric readily comparable to that of the graphs in the previous sec-
tion). The regression coefficients, the parametric assumptions concern-
ing the survival function, and the data can be combined to estimate the
probability of distress for each bank for each day of a crisis period. For
the crisis during the fall of 1930, the mean estimate is 1.74 failures per
day per thousand banks. A counterfactual—what would the hazard rate
have been during the panic in the absence of the Atlanta Fed’s inter-
vention—can be estimated by setting the indicator variables for the
panic/Federal Reserve District 6 interaction equal to zero and redoing
the calculation. The mean estimate for the no-intervention counterfac-
tual is 2.43 per thousand. Another counterfactual—what would the haz-
ard rate have been in the absence of the panic—can be estimated by
setting the indicator variables for panic and panic/Federal Reserve Dis-
trict 6 interaction equal to zero. The mean estimate for the no-panic
counterfactual is 0.11 per thousand. The difference between the coun-
terfactuals indicates the impact of the panic. The difference is 2.32 per
thousand. Compounding that figure over the 155 days of the crisis re-
veals that the panic increased the cumulative hazard for the average
bank by 36.0 percent. Similar calculations indicate the effect of the
Atlanta Fed’s expansionary policy.
Columns 2–6 in tables 5 and 6 strengthen this supposition. Column
2 adds to the explanatory variables a vector of bank characteristics. The
characteristics include the percentage of total assets composed of cash,
V. Robustness
Our conclusion remains robust to a wide variety of alterations in our
econometric framework. Parametric models employing different para-
metric assumptions, explanatory variables, and corrections for hetero-
TABLE 7
Asset Quality at Suspended Banks in Mississippi, January 1929
through March 1933
Good Problematic
Number % Number %
6th Atlanta:
1. Panic 3 25.0 9 75.0
2. Nonpanic 2 8.0 23 92.0
8th St. Louis:
3. Panic 37 55.2 30 44.8
4. Nonpanic 12 38.7 19 61.3
Source.—National Archives and Records Administration, Record Group 82. See Richardson (2006) for
details.
Note.—Rows 1 and 3 present figures for all banks suspending during October, November, and December
1930 and January, February, and March 1931. Rows 2 and 4 present figures for banks suspending operations
in all other months from January 1929 through March 1933. The rows sum to 100 percent. Columns indicate
the percentage of suspended banks in each district in each period whose assets were judged by examiners
to be good and, thus, to have been neither a primary nor a contributing cause of the suspension, and problematic
(i.e., either slow, doubtful, or worthless) and to have been either a primary or a contributing cause of the
suspension.
A final type of evidence completes the case. Three sources report the
quality of assets at failed banks. First, the Federal Reserve Board St. 6386
forms indicate examiners’ ex ante (i.e., before the suspension) assess-
ment of the quality of assets at suspended banks. Examiners reported
the quality of assets at banks to be good and neither to have been a
primary nor a contributing cause of the suspension, or problematic (i.e.,
either slow, doubtful, or worthless) and to have been either a primary
or contributing cause of the suspension. Table 7 presents this infor-
mation. It shows that the quality of assets at institutions that suspended
operations in the 8th District was better than the quality in the 6th
District. During the post-Caldwell panic, the majority of the banks that
suspended operations in the 8th District had portfolios consisting pre-
dominantly of good assets.
Second, Warburton’s study, “Deposit Guaranty in Mississippi,” pro-
vides evidence on recoveries from the assets of failed banks (Warburton
1955, 41–51, tables 11–13). From 1916 to March 1930, when Mississippi
guaranteed bank deposits, recoveries averaged just over 51.5 percent
(i.e., on average, assets with a book value of $100 yielded $51.50). Re-
coveries from the assets of banks that failed during the post-Caldwell
panic averaged 70.4 percent. Third, the 1929, 1930, and 1931 Biennial
Reports of Mississippi’s Banking Department (tables F and G) record
information on recoveries from banks in liquidation. For 39 banks that
failed during the post-Caldwell panic, data exist on (a) recoveries from
the initial sale of assets shortly following suspension (these sales were
supposed to be of assets that yielded nearly book value or better), (b)
the initial estimate of the value of the remaining assets, and (c) eventual
recoveries from sales of the remaining assets during the years 1931–33.
Fig. 6.—Comparing consequences of the banking panics in the 6th and 8th Districts.
A, Total deposits as a percentage of total deposits in June 1930. B, Total loans and discounts
as a percentage of the total in June 1930.
percent, a failure rate nearly twice that of firms in the 6th District.
Wholesale transactions also fell farther in the 8th than in the 6th District.
Variation in the reason for and location of bank failures enables us
to draw clear conclusions concerning the contraction of credit and
decline in commercial activity. The regressions reported in table 9 ex-
ploit this variation. The dependent variable is the decline in wholesale
transactions (measured as 1929 real dollars) in each county from 1929
to 1933 (measured by summing net sales for all wholesalers in each
county during the year 1929 and subtracting the same sum for the year
1933). Baseline specifications appear in columns 1 and 2. Column 1’s
sole explanatory variable is the decline in total loans outstanding from
1929 to 1933 (measured by summing loans and discounts on the balance
TABLE 8
Decline in Wholesale Trade
TABLE 9
County-Level Regressions of Commercial Activity on Credit Contraction
Dependent Variable: Decline in Net Wholesale Transactions from 1929 to 1933
VII. Discussion
The multiple sources and methods employed in the previous sections
tell a consistent tale. During the banking panic that began in December
1930, banks failed at lower rates in the 6th Federal Reserve District,
where the Atlanta Fed injected liquidity into the banking system, than
in the 8th Federal Reserve District, where the St. Louis Fed followed
the doctrine of real bills. The St. Louis Fed could have followed the
same policy as the Atlanta Fed, and if it had, bank failure rates would
have been lower, commercial lending would have remained higher, and
the contraction would not have been as severe.
The quasi-experimental structure of our study, which frees our esti-
mates from difficulties of inference that typically trouble studies of firms
in complex, changing, and endogenous economic environments,
strengthens our conclusion. The limitations of our analysis are the same
as those for any study of this type. While our methods generate a precise
and powerful result, they do so for a particular point in time and space:
an agricultural state during the initial banking panic of the 1930s. The
generalizability of our result depends on the representativeness of the
place and period under study. On this dimension, our study stands on
strong ground.
Mississippi’s banks were representative of the segments of the financial
system that bore the brunt of the contraction. Mississippi was an agri-
cultural state suffering from droughts, falling commodity prices, and
References
Bernanke, Benjamin S. 1983. “Nonmonetary Effects of the Financial Crisis in
the Propagation of the Great Depression.” A.E.R. 73 (June): 257–76.
Calomiris, Charles W., and Joseph R. Mason. 2003. “Fundamentals, Panics, and
Bank Distress during the Depression.” A.E.R. 93 (December): 1615–46.
Cecchetti, Stephen G., and Georgios Karras. 1994. “Sources of Output Fluctu-
ations during the Interwar Period: Further Evidence on the Causes of the
Great Depression.” Rev. Econ. and Statis. 76 (January): 80–102
Chandler, Lester Vernon. 1971. American Monetary Policy, 1928–1941. New York:
Harper and Row.
Christiano, Lawrence J., Roberto Motto, and Massimo Rostagno. 2003. “The
Great Depression and the Friedman-Schwartz Hypothesis.” J. Money, Credit,
and Banking 35, no. 6, pt. 2 (December): 1119–97.
Diamond, Douglas W., and Philip H. Dybvig. 1983. “Bank Runs, Deposit Insur-
ance, and Liquidity.” J.P.E. 91 (June): 401–19.